FutureLens Gregory Shutt November 20, 2008

advertisement
FutureLens
Gregory Shutt
November 20, 2008
Motivation
• Visualize tagged data
• Extract features from data
• Knowledge discovery
Data
• VAST 2007 Contest
• News stories
• SGML files tagged with different types of
entities
• Person, organization, money, date,
location
<TIMEX TYPE="DATE">
Fri Aug 15 2003
</TIMEX>
<ENAMEX TYPE="PERSON">
Jon Zwickel
</ENAMEX>
<TIMEX TYPE="DATE">
Wed Aug 6 2003
</TIMEX>
wanted to create the ultimate B.C. hot dog.
Hence the world has the
<ENAMEX TYPE="ORGANIZATION">
PNE Salmon Sausage
</ENAMEX>
These [genetically engineered] products are
absolutely safe. For the most part you wouldn't
know [if you were eating them] but the point
being that you wouldn't need to know.
<ENAMEX TYPE="PERSON">
Bryan Hurley
</ENAMEX>
,
, a new taste treat that will be unveiled when the
Pacific National Exhibition opens
<TIMEX TYPE="DATE">Saturday</TIMEX>
<TIMEX TYPE="TIME">morning</TIMEX>.
<ENAMEX TYPE="ORGANIZATION">
Monsanto
</ENAMEX>
spokesperson
"There's nothing more
<ENAMEX TYPE="LOCATION">
West Coast
</ENAMEX>
than salmon," said
<ENAMEX TYPE="PERSON">
Zwickel
</ENAMEX>
Sample Data
Data
• Nonnegative tensor factorization software
used
• NTF software output 25 group files
• Each group was described by a number of
interrelated entities and terms
Requirements
• Visualize VAST 2007 data
• Simple and easily modifiable
• Maintain UI responsiveness
• Allow viewing of individual group files
Group 9
0.3588235
0.3588235
0.3588235
0.3588235
0.3588235
0.3258677
0.2219373
0.1687334
0.1465103
0.1447117
0.1373243
0.1254398
0.1246595
Relevant
Entities
$215 million
Cruz
Darla Banks
$25-30 million
Banks
Terms
banks
fishes
tropical
trafficking
brazil
illegal
poachers
fish
Irrelevant
File: Week-ofMon-20030630.xml.txt.p.NE
Something is rotten in the tropical fish import
business and not just some dead fish. A
southern environmentalist has succeeded in
trapping poachers by conducting sting
operations in Brazil - and Darla Banks loves
doing this. She carries a concealed camera in her
handbag and secretly films illegal freshwater
fish collections, including the rare Black arwana
and Cruz' Dwarf Pearlfish.
From 340-500 million fishes are kept in
American homes (three times the total number
of dogs and cats). Trade in fish grows every
year. At least US $ 215 million in tropical fishes
are handled every year in the US . The US imports
125 million of ornamental fishes per year - US $
25-30 million /year.
File: Week-ofMon-20030818.xml.txt.p.NE
Fall is time to see bighorns in Hells Canyon Out
& About A herd of 18 bighorn sheep wanders
along the banks of the Snake River in Hells
Canyon, moseying from rock to rock, drinking
from the river, and chewing on grass. Five rams,
with massive horns curling over the sides of their
heads, are in the herd. What a sight. A lamb
bounds playfully but cautiously between the
adult animals. Bighorn numbers started to
decline as soon as the state began to be settled.
They were easy to hunt and provided food for
early miners and settlers. And they were -- and
still are -- susceptible to diseases like scabies
and pasturella, which are transmitted from
domestic sheep. As homesteaders brought in
more domestic sheep, the bighorns became sick
and died. "The bottom line is there is no danger
to domestic sheep from wild sheep," Cassirer
says. "It's only one way."
Background
• Conceptually based on FeatureLens, a
University of Maryland HCIL project
• Visualizes frequent terms and patterns in
text over time
FeatureLens
FeatureLens
• Requires MySQL server, HTTP server,
Adobe Flash enabled browser
• Written in Ruby and OpenLaszlo
• Difficult to modify
• Arbitrary data sets not loadable by end
users
• Interface responsiveness is subpar
FutureLens
• Written in Java using SWT
• Cross platform with native look and feel
• Works with SGML and raw text
• Supports tagged entities
• Allows viewing of group files
Cross platform but it
maintains the look and
feel familiar to the user
of the particular
operating system
Most Java programs do
not have this capability
Demonstration
• A demonstration of how a scenario can
quickly be visualized using NTF output and
the VAST 2007 data
Future Work
• Integrate data mining software
• Allow dynamic data sets
• Use machine learning to automate tasks
References
•
Exploring and visualizing frequent patterns in text collections with FeatureLens. http://
www.cs.umd.edu/hcil/textvis/featurelens. Visited November 2008.
•
The MONK Project Wiki. https://apps.lis.uiuc.edu/wiki/display/MONK/The+MONK+Project
+Wiki. Last edited August 2008.
•
Brett W. Bader, Michael W. Berry, and Murray Brown. Discussion tracking in Enron email using
PARAFAC. In M.W. Berry and M. Castellanos, editors, Survey of Text Mining II: Clustering,
Classification, and Retrieval, pages 147–163. Springer-Verlag, London, 2008.
•
Brett W. Bader, Michael W. Berry, and Amy N. Langville. Nonnegative matrix and tensor
factorization for discussion tracking. In A. Srivastava and M. Sahami, editors, Text Mining: Theory,
Applications, and Visualization. Chapman & Hall / CRC Press, 2008.
•
Brett W. Bader, Andrey A. Puretskiy, and Michael W. Berry. Scenario discovery using nonnegative
tensor factorization. In Jose Ruiz-Shulcloper and Walter G. Kropatsch, editors, Progress in Pattern
Recognition, Image Analysis and Applications, Proceedings of the Thirteenth Iberoamerican
Congress on Pattern Recognition, CIARP 2008, Havana, Cuba, Lecture Notes in Computer
Science (LNCS) 5197, pages 791–805. Springer-Verlag, Berlin, 2008.
•
A. Don, E. Zhelev, M. Gregory, S. Tarkan, L. Auvil, T. Clement, B. Shneiderman, and C. Plaisant.
Discovering interesting usage patterns in text collections: integrating text mining with
visualization. HCIL Technical report 2007-08, May 2007.
Download